predict_bug_fastai_w_wo_loss.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from fastai import *\n",
    "import matplotlib.pyplot as plt\n",
    "from fastai.vision import *\n",
    "from fastai.callbacks.hooks import num_features_model\n",
    "import cv2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "from fastai.callbacks.hooks import num_features_model\n",
    "from fastai.vision import create_head"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "from motion.detect_human import BBoxDataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Dataset has 1150 samples.\n",
      "Dataset has 225 samples.\n",
      "(3, 500, 500) (4,)\n"
     ]
    }
   ],
   "source": [
    "SZ = 500\n",
    "bbox_ds = BBoxDataset(\"coco/val2017_one_human.csv\",type = 'val', size = SZ)\n",
    "bbox_ds_val =  BBoxDataset(\"coco/val2017_one_human_val.csv\",type='valid', size = SZ)\n",
    "print(bbox_ds_val[0][0].shape,bbox_ds_val[0][1].shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "BS = 8\n",
    "\n",
    "train_dl = DataLoader(bbox_ds, BS)\n",
    "valid_dl = DataLoader(bbox_ds_val, BS, \n",
    "                      shuffle=False, num_workers=2, pin_memory=True)\n",
    "\n",
    "data_bunch = DataBunch(train_dl, valid_dl)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "class FlukeDetector2(nn.Module):\n",
    "    def __init__(self, arch=models.resnet18):\n",
    "        super().__init__() \n",
    "        self.cnn = create_body(arch)\n",
    "        self.head = create_head(num_features_model(self.cnn) * 2, 4)\n",
    "        \n",
    "    def forward(self, im):\n",
    "        x = self.cnn(im)\n",
    "        x = self.head(x)\n",
    "        return x.sigmoid_()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Without loss function"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn = Learner(data_bunch, FlukeDetector2(arch=models.resnet50))\n",
    "learn = learn.load('fastai_bbox_detect_humans_val3')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "preds, targs = learn.get_preds()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([1.345302, 1.258236, 2.051251, 2.114838], dtype=float32)"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "predicted_bboxes = ((preds) ).numpy()\n",
    "targets = ((targs) ).numpy()\n",
    "predicted_bboxes[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Note the result is above 1 while the model end with a sigmoid function."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## With loss function"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch.nn import L1Loss\n",
    "learn = Learner(data_bunch, FlukeDetector2(arch=models.resnet50), loss_func=L1Loss())\n",
    "learn = learn.load('fastai_bbox_detect_humans_val3')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "preds, targs = learn.get_preds()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0.296618, 0.229711, 0.71845 , 0.748978], dtype=float32)"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "predicted_bboxes = ((preds) ).numpy()\n",
    "targets = ((targs) ).numpy()\n",
    "predicted_bboxes[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Different results! "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0.548, 0.236, 0.802, 0.868], dtype=float32)"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "targets[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "and this time they seem fine what I expected."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "```text\n",
      "=== Software === \n",
      "python        : 3.7.2\n",
      "fastai        : 1.0.42\n",
      "fastprogress  : 0.1.18\n",
      "torch         : 1.0.1.post2\n",
      "torch cuda    : 10.0.130 / is available\n",
      "torch cudnn   : 7402 / is enabled\n",
      "\n",
      "=== Hardware === \n",
      "torch devices : 1\n",
      "  - gpu0      : GeForce GTX 1080\n",
      "\n",
      "=== Environment === \n",
      "platform      : Linux-4.15.0-45-generic-x86_64-with-debian-buster-sid\n",
      "distro        : #48-Ubuntu SMP Tue Jan 29 16:28:13 UTC 2019\n",
      "conda env     : Unknown\n",
      "python        : /home/tako/devtools/furry-geras/env/bin/python\n",
      "sys.path      : /home/tako/devtools/furry-geras\n",
      "/home/tako/devtools/furry-geras/env/lib/python37.zip\n",
      "/home/tako/devtools/furry-geras/env/lib/python3.7\n",
      "/home/tako/devtools/furry-geras/env/lib/python3.7/lib-dynload\n",
      "/usr/local/lib/python3.7\n",
      "\n",
      "/home/tako/devtools/furry-geras/env/lib/python3.7/site-packages\n",
      "/home/tako/devtools/furry-geras/env/lib/python3.7/site-packages/IPython/extensions\n",
      "/home/tako/.ipython\n",
      "no nvidia-smi is found\n",
      "```\n",
      "\n",
      "Please make sure to include opening/closing ``` when you paste into forums/github to make the reports appear formatted as code sections.\n",
      "\n",
      "Optional package(s) to enhance the diagnostics can be installed with:\n",
      "pip install distro\n",
      "Once installed, re-run this utility to get the additional information\n"
     ]
    }
   ],
   "source": [
    "from fastai.utils.collect_env import *\n",
    "show_install()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "env-furry-geras",
   "language": "python",
   "name": "env-furry-geras"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}